{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:3IKMYQI365UIGWU6RMUYPKOLEQ","short_pith_number":"pith:3IKMYQI3","schema_version":"1.0","canonical_sha256":"da14cc411bf768835a9e8b2987a9cb240397679a8970a97efe7ca4c03deaa6c6","source":{"kind":"arxiv","id":"2102.05800","version":3},"attestation_state":"computed","paper":{"title":"Robust Policy Gradient against Strong Data Corruption","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Wen Sun, Xiaojin Zhu, Xuezhou Zhang, Yiding Chen","submitted_at":"2021-02-11T01:48:38Z","abstract_excerpt":"We study the problem of robust reinforcement learning under adversarial corruption on both rewards and transitions. Our attack model assumes an \\textit{adaptive} adversary who can arbitrarily corrupt the reward and transition at every step within an episode, for at most $\\epsilon$-fraction of the learning episodes. Our attack model is strictly stronger than those considered in prior works. Our first result shows that no algorithm can find a better than $O(\\epsilon)$-optimal policy under our attack model. Next, we show that surprisingly the natural policy gradient (NPG) method retains a natural"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2102.05800","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2021-02-11T01:48:38Z","cross_cats_sorted":[],"title_canon_sha256":"d54545140528394a991db5fdd479aed96d50c84141f69a2ad42f0bae03e41b7a","abstract_canon_sha256":"df1e2efc5ff1facb982a9a6d7b30f4a7de820b568abb7896f8fb4b3ee80d33c3"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T02:47:04.847521Z","signature_b64":"4uF0z5xD87x3fKbw1nybCy1+LB/gKyYCMBR7SUsvDrd6YqOgjpfTMqw0RPWW6WicG47/W5THwGMcqn56FkiiAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"da14cc411bf768835a9e8b2987a9cb240397679a8970a97efe7ca4c03deaa6c6","last_reissued_at":"2026-07-05T02:47:04.847075Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T02:47:04.847075Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Robust Policy Gradient against Strong Data Corruption","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Wen Sun, Xiaojin Zhu, Xuezhou Zhang, Yiding Chen","submitted_at":"2021-02-11T01:48:38Z","abstract_excerpt":"We study the problem of robust reinforcement learning under adversarial corruption on both rewards and transitions. Our attack model assumes an \\textit{adaptive} adversary who can arbitrarily corrupt the reward and transition at every step within an episode, for at most $\\epsilon$-fraction of the learning episodes. Our attack model is strictly stronger than those considered in prior works. Our first result shows that no algorithm can find a better than $O(\\epsilon)$-optimal policy under our attack model. Next, we show that surprisingly the natural policy gradient (NPG) method retains a natural"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2102.05800","kind":"arxiv","version":3},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2102.05800/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2102.05800","created_at":"2026-07-05T02:47:04.847131+00:00"},{"alias_kind":"arxiv_version","alias_value":"2102.05800v3","created_at":"2026-07-05T02:47:04.847131+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2102.05800","created_at":"2026-07-05T02:47:04.847131+00:00"},{"alias_kind":"pith_short_12","alias_value":"3IKMYQI365UI","created_at":"2026-07-05T02:47:04.847131+00:00"},{"alias_kind":"pith_short_16","alias_value":"3IKMYQI365UIGWU6","created_at":"2026-07-05T02:47:04.847131+00:00"},{"alias_kind":"pith_short_8","alias_value":"3IKMYQI3","created_at":"2026-07-05T02:47:04.847131+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.22579","citing_title":"Stationary Robust Mean-Field Games under Model Mismatches","ref_index":103,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/3IKMYQI365UIGWU6RMUYPKOLEQ","json":"https://pith.science/pith/3IKMYQI365UIGWU6RMUYPKOLEQ.json","graph_json":"https://pith.science/api/pith-number/3IKMYQI365UIGWU6RMUYPKOLEQ/graph.json","events_json":"https://pith.science/api/pith-number/3IKMYQI365UIGWU6RMUYPKOLEQ/events.json","paper":"https://pith.science/paper/3IKMYQI3"},"agent_actions":{"view_html":"https://pith.science/pith/3IKMYQI365UIGWU6RMUYPKOLEQ","download_json":"https://pith.science/pith/3IKMYQI365UIGWU6RMUYPKOLEQ.json","view_paper":"https://pith.science/paper/3IKMYQI3","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2102.05800&json=true","fetch_graph":"https://pith.science/api/pith-number/3IKMYQI365UIGWU6RMUYPKOLEQ/graph.json","fetch_events":"https://pith.science/api/pith-number/3IKMYQI365UIGWU6RMUYPKOLEQ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/3IKMYQI365UIGWU6RMUYPKOLEQ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/3IKMYQI365UIGWU6RMUYPKOLEQ/action/storage_attestation","attest_author":"https://pith.science/pith/3IKMYQI365UIGWU6RMUYPKOLEQ/action/author_attestation","sign_citation":"https://pith.science/pith/3IKMYQI365UIGWU6RMUYPKOLEQ/action/citation_signature","submit_replication":"https://pith.science/pith/3IKMYQI365UIGWU6RMUYPKOLEQ/action/replication_record"}},"created_at":"2026-07-05T02:47:04.847131+00:00","updated_at":"2026-07-05T02:47:04.847131+00:00"}